Thin-wall part milling vibration active control method based on deep Q learning
Through deep Q learning, the milling vibration of thin-walled parts is controlled, and the problems of poor vibration damping effect and insufficient information processing capabilities caused by dependence on precise models in the prior art are solved, and adaptive control of complex processing environments is realized, which significantly improves thin Suppression effect of wall milling vibration.
Patent Information
- Application Number
- CN202411290169.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-13
- Publication Date
- 2024-12-27
AI Technical Summary
The existing thin-walled part milling vibration control methods rely too much on precise models and cannot adapt to complex and uncertain processing environments, resulting in poor vibration damping effects, insufficient controller information processing capabilities, and unable to meet high-demand vibration control.
Adopting the active control method of milling vibration of thin-walled parts based on deep Q learning, the design of state, action, reward function and neural network by establishing lateral vibration equations, piezoelectric sheet sensing and actuation equations, and training the deep Q learning controller to realize Adaptive vibration control of thin-walled parts is independent of the precise model.
It improves the adaptability and processing complexity of milling vibration control of thin-walled parts, improves the control performance of the system, can effectively suppress vibration under unknown or variable conditions, and significantly improves the milling vibration control effect.
Smart Images
Figure CN119203410A_ABST
Abstract
Description
Technical field:
[0001] The present invention belongs to the technical field of mechanical processing, and in particular relates to a method for actively controlling vibration of thin-walled part milling based on deep Q learning. Background technology:
[0002] Thin-walled parts are representative products of the advanced stage of manufacturing industry, and are widely used in many important fields of the country, such as aerospace, energy, transportation, etc. However, thin-walled parts are very prone to vibration during processing. When the processing conditions are poor and the workpiece processing parameters do not meet the specific requirements, it will lead to poor surface quality of the workpiece, tool wear and other negative effects. Therefore, it is necessary to try to avoid vibration during processing.
[0003] At present, the most commonly used thin-walled parts milling vibration control methods are mainly divided into two categories, one is passive control, and the other is active and semi-active control. Passive control is mainly through the analysis and calculation of the parameters in the workpiece processing, and then the workpiece working conditions are optimized in advance. The main methods include the optimization of processing parameters, the structural optimization of the processing system, and the application of shock absorbers or passive dampers. Active and semi-active control are mostly closed-loop systems. The system is mainly composed of a controller and an actuator. The controller monitors the vibration signal during the processing process and adjusts the output of the actuator through the algorithm set in advance by the computer; the actuator mainly receives the vibration signal from the controller and responds accordingly.
[0004] Liu, GY et al. studied thin-walled parts under milling force and designed a set of optimal tool helix angles β and tilt angles θ, which significantly suppressed machining vibration and improved machining surface quality. (Liu, Gongyu Dang, Jiaqiang Li, Chao Ming, Weiwei An, Qinglong Chen, Ming. Investigation on the vibration and machined surface quality in tilt side milling of thin-walled plates [J]. The International Journal of Advanced Manufacturing Technology, 2019, 103 (5a8).) However, the optimal cutting parameters are different under different working conditions and are overly dependent on the precise model of the controlled object, resulting in poor vibration reduction effect. Wang Jian et al. proposed an active damping device with direct velocity feedback, meshed and modal simulation analysis of thin-walled parts using finite element software, and verified the vibration suppression effect of the active damping device by building a test bench. (Wang Jian, He Lidong, Yang Jianjiang. Research on milling vibration control of thin-walled parts based on active damping technology [J]. Mechanical and Electrical Engineering, 2022, 39(02): 262-268.) However, the algorithm used in the process of active vibration suppression control is relatively traditional. When controlling thin-walled parts under milling force, it cannot adapt to the complexity and uncertainty of the object and has poor information processing capabilities. It cannot learn the information obtained from the external environment, and it is difficult to improve the control performance of the system and meet higher control requirements. KJKalinski et al. used optimal control to apply milling vibration suppression to milling thin-walled parts. The acceleration sensor was arranged on the surface of the workpiece to monitor the vibration response of the workpiece processing. The energy performance index was used as the control strategy of the controller to provide the corresponding control signal to the piezoelectric actuator and use the piezoelectric actuator to control the vibration. (Kaliński KJ, Galewski M A. Vibration surveillance supported by Hardware-In-the-Loop Simulation in milling flexible workpieces[J]. Mechatronics, 2014, 24(8): 1071-1082.) However, the optimal controller used relies on the accurate model of the controlled object and has poor optimization capability, and has no obvious advantage when dealing with complex and uncertain systems.
[0005] At present, the methods for controlling vibration in thin-walled parts milling have the disadvantages of over-reliance on the precise model of the controlled object, resulting in poor vibration reduction effect, inability to handle the complexity and uncertainty of thin-walled parts milling, and poor information processing ability and performance of the controller in the milling process. The active deep Q learning controller for thin-walled parts milling vibration can solve these disadvantages and has the following advantages: adapting to the complexity and uncertainty of thin-walled parts in the milling process and better handling the vibration control task of thin-walled parts milling; having a learning function, learning through a large number of thin-walled parts milling vibration signals obtained from the thin-walled parts milling process, constantly accumulating knowledge, and improving the active vibration control performance of thin-walled parts milling system; not relying on the precise model of thin-walled parts in the milling process, and adapting to the situation where the thin-walled parts model is unknown or changes greatly in the milling process; using the data generated in the milling process for optimization, by optimizing the control parameters of the deep Q learning controller, the vibration control performance of thin-walled parts milling is improved, and the active vibration control of thin-walled parts milling is realized. Based on this, this paper proposes an active vibration control method for thin-walled parts milling based on deep Q learning. The current methods for vibration control in thin-walled part milling have the disadvantages of over-reliance on the precise model of the controlled object, resulting in poor vibration reduction effect, inability to handle the complexity and uncertainty in the thin-walled part milling process, and poor information processing capability and performance of the controller during the milling process. Summary of the invention:
[0006] The present invention proposes a method for active vibration control of thin-walled parts milling based on deep Q learning, which provides theoretical support and technical guidance for active vibration control of thin-walled parts milling.
[0007] The technical solution adopted by the present invention is: a method for actively controlling vibration of thin-walled parts milling based on deep Q learning, and the specific steps are:
[0008] Step 1: Establish the lateral vibration equation of thin-walled parts; establish the sensing and actuation equation of the piezoelectric piece; establish the vibration control equation of thin-walled parts milling; use the orthogonality of the vibration mode function to establish the modal equation of thin-walled parts milling vibration control; use the state variable to establish the state space equation of thin-walled parts milling vibration control; use COMSOL finite element software to verify the model and obtain the natural frequency of the thin-walled parts. Under the action of milling force and piezoelectric piece, the vibration control equation of thin-walled parts milling is:
[0009]
[0010] Among them, f(x,y,t) is the milling force, q t is the control force of the piezoelectric piece, D is the bending stiffness of the thin-walled part (N·m), w(x,y,t) is the vibration displacement of thin-walled parts during milling. is the Laplace operator, we have E is the elastic modulus of the thin-walled part (N / m 2), μ is Poisson’s ratio, ρ is the density of the thin-walled part (kg / m3), h is the thickness of the thin-walled part (m),
[0011] Step 2: Establish a Matlab / Simulink thin-walled parts milling vibration environment model; design the state, action, and reward function of thin-walled parts; establish the active control framework of the thin-walled parts milling vibration active deep Q learning controller; design the learning rate, neural network design, design discount factor, experience pool, sampling time, maximum training rounds, and maximum number of steps for a single training; establish the process of the thin-walled parts milling vibration active deep Q learning controller; train the thin-walled parts milling vibration active deep Q learning controller to obtain the parameters of the milling vibration active deep Q learning controller.
[0012] Step three: During the milling process, use the trained thin-walled workpiece milling vibration active deep Q learning controller to control the vibration of the thin-walled workpiece. The specific control principle is that during the milling process, the piezoelectric sensor collects the thin-walled workpiece milling vibration displacement signal and transmits it to the thin-walled workpiece milling vibration active deep Q learning controller. When the controller is trained, it outputs a voltage, which is transmitted to the piezoelectric piece through the power amplifier. The piezoelectric piece converts electrical energy into mechanical energy, controls the thin-walled workpiece under the action of milling force, and realizes active control of thin-walled workpiece milling vibration.
[0013] In step 2, when actively controlling the milling vibration of thin-walled parts under milling, the state, action, reward function, learning rate, neural network and discount factor are designed.
[0014] (1) Design status
[0015] During the vibration process, the state of the thin-walled part under milling is designed, and the displacement and velocity of the thin-walled part milling vibration are selected. Since reinforcement learning requires the state to be unique, two physical quantities must be used to define the state of the control object during the vibration process. The control object is a thin-walled part under the milling force, and the modal coordinates of the thin-walled part milling vibration are designed as the displacement-velocity pair (η t ,η t ′), uniquely determines the position and velocity of the controlled object at time t. Therefore, the state space of the designed thin-walled part milling vibration is S = (X, V).
[0016] (2) Design Action
[0017] When actively controlling the milling vibration of thin-walled parts, different voltages u are output through the piezoelectric sheet for control. Therefore, the design voltage u is the action a, and the action space is A=U.
[0018] The intelligent agent applies action a to the thin-walled part under the milling force, so that the thin-walled part under the milling force enters the next state. The intelligent agent takes the next action based on the information of the reward value obtained from the state action.
[0019] (3) Designing the Reward Function
[0020] During the milling process, it is hoped that the vibration displacement and speed of thin-walled parts milling are as small as possible. Based on this, the design reward function is:
[0021]
[0022] Where η is the vibration modal coordinate of thin-walled parts milling, is the first-order derivative of the modal coordinate milling vibration of thin-walled parts. k1 and k2 are proportional coefficients.
[0023] (4) Design learning rate
[0024] If the learning rate is designed to be too small, the convergence process will become very slow, and it is easy to overfit and fall into the local optimal solution. If the learning rate is set too large, the gradient will oscillate and not converge.
[0025] (5) Design discount factor
[0026] The discount factor is a constant between 0 and 1. The larger the discount factor, the greater the future reward, and vice versa.
[0027] Furthermore, in step 3, during the actual cutting process, the vibration response in the y direction is greater than the vibration response in the x direction. In order to simplify the model, only the milling force acting on the thin-walled part in the y direction is considered. The transfer function model of the thin-walled part is imported into the simulink software, and the milling force is used as the input signal and imported into the thin-walled part for active control of thin-walled part milling vibration by deep Q learning.
[0028] The present invention provides an active vibration control method for thin-walled parts milling based on deep Q learning. The current method for vibration control of thin-walled parts milling has the disadvantages of over-reliance on the precise model of the controlled object, resulting in poor vibration reduction effect, inability to handle the complexity and uncertainty in the milling process of thin-walled parts, and poor information processing capability and performance of the controller during the milling process. The active deep Q learning controller for vibration of thin-walled parts milling can solve these shortcomings. Based on this, this paper proposes an active vibration control method for thin-walled parts milling based on deep Q learning.
[0029] The present invention has the following beneficial effects:
[0030] 1. By using piezoelectric sheets as actuators, it has the characteristics of high sensitivity and lightness, and can be used in the milling of thin-walled parts.
[0031] 2. Design an active deep Q-learning controller for thin-walled parts milling vibration to adapt to the complexity and uncertainty of thin-walled parts during the milling process and better handle the milling vibration control task of thin-walled parts.
[0032] 3. Design an active deep Q learning controller for thin-walled part milling vibration, which has a learning function. It learns through a large number of thin-walled part milling vibration signals obtained from the thin-walled part milling environment, continuously accumulates knowledge, and greatly improves the control performance of the system.
[0033] 4. Design an active deep Q-learning controller for milling vibration of thin-walled parts, which does not rely on the precise model of the thin-walled parts during milling and can handle the situation where the model of the thin-walled parts is unknown or changes greatly during milling;
[0034] 5. Design an active deep Q-learning controller for the milling vibration of thin-walled parts, use the data generated during the milling process for optimization, and improve the milling vibration control performance of thin-walled parts by optimizing the control parameters of the deep Q-learning controller to achieve active control of milling vibration of thin-walled parts.
[0035] 6. Through comprehensive analysis of the active vibration control of thin-walled parts milling based on deep Q learning, the vibration state of thin-walled parts milling in the actual milling process can be more accurately predicted and evaluated, providing a reference for the active vibration control of thin-walled parts milling. Description of the drawings:
[0036] Figure 1 It is a flowchart of the present invention;
[0037] Figure 2 is a COMSOL finite element model of the thin-walled part described in the present invention;
[0038] Figure 3 It is the structure composition of the thin-walled part milling vibration active control system based on deep Q learning described in the present invention;
[0039] Figure 4 The present invention is a thin-walled part milling vibration active deep Q learning model based on Matlab / Simulink;
[0040] Figure 5 The thin-walled parts milling environment model based on Matlab / Simulink described in the present invention;
[0041] Figure 6 It is a framework diagram of the thin-walled part milling vibration active control system based on deep Q learning described in the present invention;
[0042] Figure 7 It is a flow chart of active vibration control of thin-walled parts milling based on deep Q learning according to the present invention;
[0043] Figure 8is the milling force diagram of the present invention;
[0044] Fig. 9 This is a vibration control diagram for thin-walled part milling based on three control methods described in the present invention. Specific implementation method:
[0045] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0046] Specific embodiments of the present invention:
[0047] A method for active vibration control of thin-walled parts milling based on deep Q learning, the specific steps are as follows:
[0048] Step 1: Establish the lateral vibration equation of thin-walled parts. The mid-surface of a rectangular thin-walled part of equal thickness coincides with the (x, y) plane of the coordinate system. The z-axis is the vertical direction of the thin-walled part surface. The displacement of a point on the mid-surface of the thin-walled part at a certain time along the z-axis direction is. Based on the Kirchhoff-love hypothesis, the free vibration equation of a two-dimensional rectangular plate is established:
[0049]
[0050] Where D is the bending stiffness of the thin-walled part (N·m), w(x,y,t) is the vibration displacement of the thin-walled part; is the Laplace operator, we have E is the elastic modulus of the thin-walled part (N / m 2 ), μ is Poisson’s ratio, ρ is the density of the thin-walled part (kg / m3), and h is the thickness of the thin-walled part (m).
[0051] The lateral displacement vibration mode of the cantilever thin-walled part is equivalent to the combination of the vibration mode functions of the cantilever boundary beam and the free boundary beam.
[0052]
[0053] Combined with the orthogonality of formula (4), formula (3) is simplified to:
[0054]
[0055] in, ω ij is the (i, j)th natural frequency of the thin-walled part.
[0056] Using the positive piezoelectric effect of the piezoelectric sheet, the piezoelectric sensing equation is established:
[0057]
[0058] y(t)=q(t) / C p (7)
[0059]
[0060] Among them, (x1, y1) is the lower left coordinate of the piezoelectric sheet, (x2, y2) is the upper right coordinate of the piezoelectric sheet, P(x, y) is the polarization direction of the piezoelectric sheet at this point, which is ±1 respectively, F(x, y) is 1 or 0 respectively according to whether the piezoelectric sheet at this point is covered by the electrode, e 31 、e 32 、e 36 is the piezoelectric stress constant coefficient of the piezoelectric sheet, w is the milling vibration displacement of the thin-walled part, h b is the thickness of the thin-walled part, h p is the thickness of the piezoelectric film, C p is the capacitance of the piezoelectric film.
[0061] Using the inverse piezoelectric effect of the piezoelectric sheet, the piezoelectric actuation equation is established:
[0062] m x =C0ε pe3 [H(x-x1)-H(x-x2)][H(y-y1)-H(y-y2)] (9)
[0063] m y =C0ε pe1 [H(x-x1)-H(x-x2)][H(y-y1)-H(y-y2)] (10)
[0064] in, d 31 d 33 is the piezoelectric coefficient of the piezoelectric film, h p is the thickness of the piezoelectric film, V n is the voltage applied to each piezoelectric patch, (x1, y1) is the lower left coordinate of the piezoelectric patch, (x2, y2) is the upper right coordinate of the piezoelectric patch, H(x) is the step function, Ep is the elastic modulus of the piezoelectric patch, v p is the Poisson's ratio of the piezoelectric patch, h b is the thickness of the thin-walled part, h p is the thickness of the piezoelectric film.
[0065] Considering the damping situation and the thin-walled part being subjected to the milling force, the milling force f(x f ,y f ,t) is:
[0066]
[0067] Among them, F ij (t) is the generalized force, ζ ij is the damping ratio,
[0068] The dynamic equation of thin-walled parts is established. Under the action of milling force and piezoelectric sheet, the control equation of milling vibration motion of thin-walled parts is:
[0069]
[0070] Among them, f(x,y,t) is the milling force, q t is the control force of the piezoelectric piece, D is the bending stiffness of the thin-walled part (N·m), w(x,y,t) is the vibration displacement of thin-walled parts during milling. is the Laplace operator, we have E is the elastic modulus of the thin-walled part (N / m 2 ), μ is Poisson’s ratio, ρ is the density of the thin-walled part (kg / m3), h is the thickness of the thin-walled part (m),
[0071]
[0072] Change formula (12) into:
[0073]
[0074] in, (x1,y1) and (x2,y2) are the lower left and upper right coordinates of the effective area of the piezoelectric patch respectively.
[0075] Multiply both sides of the above equation by the vibration mode function W of the thin-walled part jk (x, y), and integrate the entire thin-walled part. Using the orthogonality of the vibration mode function, the milling vibration equation of the thin-walled part in the modal coordinates is obtained:
[0076]
[0077] The state space equation for active control of milling vibration of thin-walled parts is established. The state variables are introduced into the milling vibration equation (14) of thin-walled parts in modal coordinates:
[0078] x=[η 11 ,…,η mn ,η1′1,…,η′ mn ] T (15)
[0079] Since the mass of the piezoelectric piece is negligible relative to the mass of the thin-walled component, its influence on the mode is not considered, and its expression is:
[0080]
[0081] Among them, A1 is the state matrix, B1 is the input voltage control matrix, B2 is the milling force matrix, C1 is the sensing matrix, y(t) is the output voltage of the piezoelectric sensor, and u(t) is the input voltage of the piezoelectric sensor.
[0082] in, A1∈R 2(m×n)×2(m×n) ,Ω=diag(ω mn ),
[0083] B1=[0,…,0,P 11 ,…,P mn ] T , C1=[S 11 ,…,S mn ,0,…,0]
[0084]
[0085]
[0086] Step 2: Study the active control strategy of thin-walled parts milling vibration based on deep Q learning, design the active deep Q learning controller for thin-walled parts milling vibration, and design the state, action, reward function, learning rate, neural network, discount factor, experience pool, sampling time, maximum training rounds, and maximum number of steps for a single training when performing deep Q learning active control of thin-walled parts milling vibration.
[0087] (1) Design status
[0088] In the milling vibration process, the state of the thin-walled part under milling is designed, and the displacement and velocity of the thin-walled part milling vibration are selected. Since reinforcement learning requires the state to be unique, two physical quantities must be used to define the state of the control object in the milling vibration process. The control object is the thin-walled part under the milling force, and the milling vibration modal coordinates of the thin-walled part are designed as the displacement-velocity pair (η t ,η t ′), uniquely determines the position and velocity of the controlled object at time t. Therefore, the state space of the designed thin-walled part milling vibration is S = (X, V).
[0089] (2) Design Action
[0090] When controlling thin-walled parts under milling force, different voltages u are output through the piezoelectric sheet for control. Therefore, the design voltage u is action a, and the action space is A=U.
[0091] The agent applies action a to the thin-walled part under the milling force, causing the milling vibration of the thin-walled part to enter the next state. The agent takes the next action based on the information of the reward value obtained from the state action.
[0092] (3) Designing the Reward Function
[0093] During the milling process, it is hoped that the vibration displacement and speed of thin-walled parts milling are as small as possible. Based on this, the design reward function is:
[0094]
[0095] Where, η is the vibration modal coordinate of thin-walled parts milling, is the first-order derivative of the vibration modal coordinates of thin-walled parts milling. k1 and k2 are proportional coefficients.
[0096] (4) Design learning rate
[0097] If the learning rate is designed to be too small, the convergence process will become very slow, and it is easy to overfit and fall into the local optimal solution. If the learning rate is set too large, the gradient will oscillate and not converge.
[0098] (5) Design discount factor
[0099] The discount factor is a constant between 0 and 1. The larger the discount factor, the greater the future reward, and vice versa.
[0100] Design process of active deep Q learning controller for vibration in thin-walled part milling:
[0101] (1): For the current evaluation network Q ω The parameters ω in (s,a) are initialized.
[0102] (2): Assign the parameter ω to the target evaluation network Omega - .
[0103] (3): Initialize the experience pool and replay pool R.
[0104] (4): Obtain the initial state s1 of the milling vibration of thin-walled parts.
[0105] (5): According to the current evaluation network Q ω (s,a) uses the ε-greedy strategy to obtain the main force a t .
[0106] (6): The main force a of the piezoelectric film t Input to Simulink environment to execute and obtain the current return r t and the state s of the milling vibration of thin-walled parts at the next moment t+1 .
[0107] (7): The four-tuple (s t ,a t ,r t ,s t+1 ) is stored in the experience replay pool R.
[0108] (8): If there is enough data in the experience pool and replay pool R, sample from R to obtain N data {(s i ,a i ,r i ,s i+1 )} i＝1,…,N .
[0109] (9): Using target evaluation network calculate
[0110] (10): Use SGD to minimize the objective loss function Perform back propagation to update the current network Q ω (s,a), and obtain the parameter ω.
[0111] (11): According to the parameters obtained in step (10), the target evaluation network is given every c steps. The parameter ω - .
[0112] Repeat the above steps to complete the parameter update in each round, and then start the next round of training until the learning process converges, completing the entire thin-walled parts milling vibration active deep Q learning agent learning process.
[0113] Step 3: During the milling process, the trained thin-walled part milling vibration active deep Q learning controller is used to control the vibration of the thin-walled part. The principle of the thin-walled part milling vibration active deep Q learning control is as follows: Figure 6 As shown. The specific control principle is that during the milling process, the piezoelectric sensor collects the milling vibration displacement signal of the thin-walled part and transmits it to the thin-walled part milling vibration active deep Q learning controller. When the controller is trained, it outputs a voltage, which is transmitted to the piezoelectric plate through the power amplifier. The piezoelectric plate converts electrical energy into mechanical energy, controls the thin-walled parts under the milling force, and realizes the active control of the milling vibration of the thin-walled parts. The parameter method of the present invention is explained more clearly in conjunction with a specific example. The thin-walled plate is made of aluminum alloy material with a density of ρ=2730kg / m 3 , elastic modulus E = 76.5 GPa, Poisson's ratio v = 0.27, thin-walled parts milling vibration active deep Q learning controller parameters: learning rate is 10 -5 , a neural network with 2 hidden layers and an experience pool size of 10 6, the sampling time is 0.01s, the maximum training round is 500, the maximum number of steps for a single training is 400, and the design discount factor is 0.99. Under the milling force signal, when the thin-walled part is not controlled, the maximum vibration displacement is 2.109×10 -3 m. Under the milling force signal, when the thin-walled part is passively controlled, the maximum vibration displacement is 1.459×10 -3 m, and the vibration suppression rate is 30.82% respectively. Under the milling force signal, when the thin-walled part is subjected to the fuzzy PID control in the traditional vibration active control, the maximum vibration displacement is 9.618×10 -4 m, and the vibration suppression rate was 54.39% respectively. Under the milling force signal, the maximum vibration displacement of the thin-walled part was reduced to 4.864×10 -4 m, and the vibration suppression rate was 77.79% respectively. Therefore, in the comparison of these three controllers, the deep Q learning controller still has the best vibration suppression effect on the vibration of thin-walled parts under the action of milling force, which solves the shortcomings of the current thin-walled parts milling vibration control method, which has excessive reliance on the precise model of the controlled object, resulting in poor vibration reduction effect, inability to handle the complexity and uncertainty of the thin-walled parts milling process, and poor information processing ability and performance of the controller during the milling process. The results are shown in Fig. 9 .
[0114] The above description is only a preferred embodiment of the present invention and does not impose any formal limitation on the structure of the present invention. Any simple modification, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are within the protection scope of the present invention.
Claims
1. An active vibration control method for thin-walled parts milling based on deep Q learning, characterized in that: The Method The method includes the following steps: S1: establishing a lateral vibration equation for thin-walled parts; establishing a piezoelectric sensing equation and a piezoelectric actuation equation; establishing a vibration control equation for thin-walled parts milling; establishing a modal equation for vibration control for thin-walled parts milling using the orthogonality of vibration mode functions; establishing a state space equation for vibration control for thin-walled parts milling using state variables; and verifying the model using COMSOL finite element software to obtain the first seven natural frequencies of the thin-walled parts. S2: Establish a Matlab / Simulink thin-walled parts milling vibration environment model; design the state, action, and reward function of thin-walled parts; establish the active control framework of the thin-walled parts milling vibration active deep Q learning controller; design the learning rate, neural network design, design discount factor, experience pool, sampling time, maximum training rounds, and maximum number of steps for a single training; The process of establishing an active deep Q learning controller for milling vibration of thin-walled parts is carried out; the active deep Q learning controller for milling vibration of thin-walled parts is trained to obtain the parameters of the active deep Q learning controller for milling vibration. S3: During the milling process, the trained active deep Q learning controller for thin-walled parts milling vibration is used to actively control the milling vibration and realize active control of thin-walled parts milling vibration.
2. The method for active vibration control of thin-walled workpiece milling based on deep Q learning as claimed in claim 1, characterized in that: The step S1 specifically includes: S11: The Kirchhoff-Love theory is used to establish the lateral vibration equation of thin-walled parts: Where D is the bending stiffness of the thin-walled part (N·m), w(x,y,t) is the vertical displacement of the thin-walled part; is the Laplace operator, we have E is the elastic modulus of the thin-walled part (N / m 2 ), μ is Poisson's ratio, ρ is the density of the thin-walled part (kg / m3), and h is the thickness of the thin-walled part (m). S12: Using the positive piezoelectric effect of the piezoelectric sheet, establish the piezoelectric sensing equation: y(t)=q(t) / C p (3) Among them, (x1, y1) is the lower left coordinate of the piezoelectric sheet, (x2, y2) is the upper right coordinate of the piezoelectric sheet, P(x, y) is the polarization direction of the piezoelectric sheet at this point, which is ±1 respectively, F(x, y) is 1 or 0 respectively according to whether the piezoelectric sheet at this point is covered by the electrode, e 31 、e 32 、e 36 is the piezoelectric stress constant coefficient of the piezoelectric sheet, w is the milling vibration displacement of the thin-walled part, h b is the thickness of the thin-walled part, h p is the thickness of the piezoelectric film, C p is the capacitance of the piezoelectric film. S13: Using the inverse piezoelectric effect of the piezoelectric sheet, the piezoelectric actuation equation is established: m x =C0ε pe3 [H(x-x1)-H(x-x2)][H(y-y1)-H(y-y2)] (5) m y =C0ε pe1 [H(x-x1)-H(x-x2)][H(y-y1)-H(y-y2)] (6) in, d 31 ,d 33 is the piezoelectric coefficient of the piezoelectric film, h p is the thickness of the piezoelectric film, V n is the voltage applied to each piezoelectric patch, (x1, y1) is the lower left coordinate of the piezoelectric patch, (x2, y2) is the upper right coordinate of the piezoelectric patch, H(x) is the step function, E p is the elastic modulus of the piezoelectric film, v p is the Poisson's ratio of the piezoelectric patch, h b is the thickness of the thin-walled part, h p is the thickness of the piezoelectric film. S14: Under the action of milling force and piezoelectric sheet, the milling vibration control equation of thin-walled parts is established: Among them, f(x,y,t) is the milling force, q t is the control force of the piezoelectric piece, D is the bending stiffness of the thin-walled part (N·m), w(x,y,t) is the vibration displacement of thin-walled parts during milling. is the Laplace operator, we have E is the elastic modulus of the thin-walled part (N / m 2 ), μ is Poisson's ratio, ρ is the density of the thin-walled part (kg / m3), h is the thickness of the thin-walled part (m), S15: Using the orthogonality of the vibration mode function, the vibration control equation for thin-walled part milling in modal coordinates is obtained: Using the state variable x = [η 11 ,…,η mn ,η′ 11 ,…,η′ mn ] T , the state space equation of vibration control in thin-walled part milling is obtained: Where A1 is the state matrix, B1 is the input voltage control matrix, B2 is the milling force matrix, C1 is the sensing matrix, y(t) is the output voltage of the piezoelectric sensor, and u(t) is the input voltage of the piezoelectric sensor. in, A1∈R 2(m×n)×2(m×n) ,Ω=diag(ω mn ), B1=[0,…,0,P 11 ,…,P mn ] T ,C1=[S 11 ,…,S mn ,0,…,0] The step S2 specifically includes: S21: The state is the modal coordinate displacement-velocity pair (η t ,η′ t ); using piezoelectric sheets to output different voltages u to actively control the vibration of thin-walled parts during milling, the voltage u is the action a, and the action space is A=U; the reward function is: Where η is the modal coordinate of the milling vibration of thin-walled parts, is the first-order derivative of the vibration modal coordinates of thin-walled parts milling. k1 and k2 are proportional coefficients. S22: The thin-walled workpiece milling vibration active deep Q learning controller control process, the steps are as follows: (1): For the current evaluation network Q ω The parameters ω in (s,a) are initialized. (2): Assign the parameter ω to the target evaluation network (3): Initialize the experience pool and replay pool R. (4): Obtain the initial state s1 of the milling vibration of thin-walled parts. (5): According to the current evaluation network Q ω (s,a) uses the ε-greedy strategy to obtain the main force a t . (6): The main force a of the piezoelectric film t Input to Simulink environment to execute and obtain the current return r t and the state s of the milling vibration of thin-walled parts at the next moment t+1 . (7): The four-tuple (s t ,a t ,r t ,s t+1 ) is stored in the experience replay pool R. (8): If there is enough data in the experience pool and replay pool R, sample from R to obtain N data {(s i ,a i ,r i ,s i+1 )} i＝1,…,N . (9): Using target evaluation network calculate (10): Use SGD to minimize the objective loss function Perform back propagation to update the current network Q ω (s,a), and obtain the parameter ω. (11): According to the parameters obtained in step (10), the target evaluation network is given every c steps. The parameter ω - . Repeat the above steps to complete the parameter update in each round, and then start the next round of training until the learning process converges, completing the entire thin-walled parts milling vibration active deep Q learning agent learning process. The step S3 specifically includes: S31: Use a displacement sensor to collect the displacement signal of the milling force acting on the thin-walled workpiece, transmit it to the thin-walled workpiece milling vibration active deep Q learning controller, and transmit it to the piezoelectric piece through the power amplifier. The piezoelectric piece converts electrical energy into mechanical energy to achieve active control of thin-walled workpiece milling vibration.